review - SKILL.md Agent Skill

name: review description: Run a code review via the refactor pipeline in review-only mode, then produce MD + HTML reports from the review findings and synthesized fix plans. argument-hint: [changes|folder|file|files|buffer] [target] [--verbose] allowed-tools: Read, Glob, Bash, Write, Skill, mcp__plugin_solid-coder_pipeline__get_output_path tools: Read, Glob, Bash, Write, Skill, mcp__plugin_solid-coder_pipeline__get_output_path user-invocable: true

Review Orchestrator (thin wrapper)

/review runs /refactor --review-only then renders MD + HTML reports. All review/synthesis logic lives in /refactor; this skill owns only the wiring and reporting.

Input

OUTPUT_ROOT: ~/.solid-coder/{project-slug}/review-<YYYYMMDDhhmmss> (computed via MCP tool)

Phase 1: Stage output root

1.1 Call mcp__plugin_solid-coder_pipeline__get_output_path with {operation: "review"} → store result as OUTPUT_ROOT

Phase 2: Run refactor in review-only mode

2.1 Use skill solid-coder:refactor with arguments: $ARGUMENTS --review-only --output-root {OUTPUT_ROOT}
- Refactor runs: discover → prepare → review → validate → synthesize, then stops.
2.2 Wait for refactor to complete. If it failed, stop and report the error.
2.3 Verify {OUTPUT_ROOT}/1/refactor-log.json exists. If missing, stop and report.
2.4 Verify {OUTPUT_ROOT}/1/by-file/ is non-empty (review produced findings). If empty, stop and report.

Phase 3: Generate reports

3.1 Run: ! python3 ${CLAUDE_PLUGIN_ROOT}/mcp-server/gateway.py generate_report --data-dir {OUTPUT_ROOT} --report-dir {OUTPUT_ROOT}
- The script aggregates across ALL iteration subdirs (1/, 2/, ...) and produces one deduplicated report organized by file.
3.2 Parse JSON output. If success is false → stop and report the error.
3.3 Report both paths: {OUTPUT_ROOT}/report.md and {OUTPUT_ROOT}/report.html.

Constraints

Do NOT modify any source code files.
Do NOT invoke any review/synthesis logic directly — delegate to /refactor.
If /refactor reports all-compliant and no synthesized plans exist, still generate reports from review findings (MD + HTML will reflect a clean state).